That shiny AI tool won't fix anything if your processes are a mess.
I get it. The headlines are full of these wild AI breakthroughs. Every other day brings a new tool, promising to "fundamentally change" your business. For small and medium-sized enterprises (SMEs), it feels like a gold rush; everyone wants a piece. My clients, from Mumbai to across the globe – local service businesses or growing e-commerce platforms like HomeGlazer – frequently ask about AI. They absolutely should.
My 13+ years, from leading design at big agencies like Performics.Convonix for nearly a decade, to CPO at a startup like Swimwell, and now building full-stack solutions with Hamzio, reveal a clear pattern: the biggest "AI wins" for SMEs aren't about magical, sentient bots. Not really. They're about good old-fashioned plumbing.
Think about building a house. You wouldn't just erect the frame, then sprinkle some "smart home" dust on it, hoping water flows when you turn on a tap. No way. You need pipes. You need valves. A reliable water source and drainage system are essential.
AI, especially Large Language Models (LLMs), is like a fancy, intelligent faucet. It's amazing at what it does, certainly, but it still needs connection to that water supply and drainage to actually do anything useful.
What people call "AI problems" are usually just bad plumbing.
That's the trap. Most teams buy AI tools, get super excited, then wonder why nothing really moves the needle. It's because their underlying processes are a mess. Data lives in silos, scattered everywhere.
Their systems don't talk. You can have the smartest AI on the planet, but if it can't access information at the right time, or pass its output to the next step, it just sits there. Useless.
This is where my experience bridging design and engineering becomes vital. My work transcends mere user flow design. I think about the entire system: where data originates, how it moves, its destination, and what happens at each step. My background in Psychology (BA) and UX post-grad from UT Austin taught me to observe how people actually behave with systems, not just how we wish they would. This is my edge.
So, when a client asks, "Hamza, we need AI for customer support," my initial query isn't about the LLM. I ask: "Where does your customer data live? How do new tickets arrive? What's the existing resolution process?" We're talking about pipes, not just the faucet itself. It's fundamental.
Let's talk about the real muscle: Webhooks, APIs, and some smart wiring.
The true backbone of practical automation for SMEs is connecting systems. This is crucial. It's about webhooks triggering actions when something happens – a new lead, an order update. It's about APIs exchanging data between your CRM and your inventory system. Logic flows then decide what happens next.
This is the "plumbing" I'm always working on with tools like n8n, Supabase, and Next.js. I'm not just building sleek interfaces in React; I'm building the entire operational backend that makes everything operate smoothly.
Let me share a recent example. Last month, I worked with a small manufacturing client, a business somewhat similar to an analogue schematic tooling company I once built for. They desperately wanted to streamline lead capture. Leads arrived from their website, WhatsApp, even phone calls – all ending up in disparate places. Their sales team struggled with manual data entry, follow-ups wildly inconsistent. You know the drill.
We planned it. The "AI" part they imagined was some magical entity, qualifying leads and sending personalized emails. My job was to show them the real steps, the plumbing: First, collect all leads into a single, structured database; we used Supabase. Then, set up a webhook that activates with each new lead, triggering an n8n workflow.
Inside n8n, that's where the real orchestration began. We connected it to their website's form, a custom Telegram bot for WhatsApp leads, and a simple manual entry form for call leads. The workflow standardized the lead data. Next, it checked existing contacts. Only then did we integrate an LLM. It wasn't magic.
This LLM's job wasn't to manage everything. It did a quick sentiment analysis on the lead's initial message or classified their industry based on keywords. Based on its output and other lead data, n8n then routed the lead to the correct sales agent in their CRM. It also updated a shared Google Sheet and triggered a personalized email (generated by the LLM, but sent by a reliable email service, not the LLM itself) or a WhatsApp message via the Telegram bot. Seamless integration.
This isn't "AI magic." No AI magic. It's structured data, event-driven triggers, and logical flows. The LLM was a smart component, a good sensor and processor, within a well-engineered system. This made the system smarter, faster, and more effective, but it would have been useless without the underlying connections, without the pipes.
Alright, so how do you actually start building these pipes?
Avoid the immediate "AI solution." Take a step back. Map your current processes. Where does data originate? Where does it flow? What are the manual bottlenecks? What systems are you already using?
Once you understand your existing plumbing, you can identify where a webhook connects systems, an API call fetches data, and a low-code tool like n8n manages a complex workflow. That's real automation.
LLMs and other AI tools become immensely powerful, integrated into robust plumbing. They become intelligent filters, smart classifiers, and personalized content generators that truly enhance existing workflows. But without those foundational pipes and connectors, you're just holding a fancy faucet, still thirsty. Build the pipes first.
You need to build a system that reliably moves information. Then, add intelligence. That's what I help businesses achieve.
